BNJ-Tutorial-20031208

BNJ-Tutorial-20031208

-

English
14 Pages
Read
Download
Downloading requires you to have access to the YouScribe library
Learn all about the services we offer

Description

BNJ 2.03aBeginnerDeveloper TutorialChris H. Meyer(revised by William H. Hsu)Kansas State UniversityKDD Laboratoryhttp://www.kddresearch.orgbndev.sourceforge.netContentsn Introductionn Inference Tutorialn Learning Tutorialn Coding the Wizardshttp://bndev.sourceforge.net1BNJ 2.0a Toolsn Offers many new tools in GUI formn This lecture will focus on the Inference and Learning Wizardsn We will also look at components such as evidence, CPT tables, and algorithms behind learninghttp://bndev.sourceforge.netContentsn Introductionn Inference Tutorialn Learning Tutorialn Coding the Wizardshttp://bndev.sourceforge.net2Starting the Inference Wizard (1)n Select Tools ?Inference Wizardhttp://bndev.sourceforge.netStarting the Inference Wizard (2)n Loadexisting networkorGUI networkn You may also select to have an evidence file presenthttp://bndev.sourceforge.net3Using the Inference Wizard (1)n Exact Inference Methods¤LS / Junction Tree¤Variable Elimination(elimbel)¤Loop Cutset Conditioning¤Pearl’s Propagation(tree only)http://bndev.sourceforge.netUsing the Inference Wizard (2)n L-S Algorithm contains 2 main steps:¤Creates a tree of cliques (junction tree) from the Bayesian Network¤Computes probability of cliques, then single-node properties are formed based on probability of cliqueshttp://bndev.sourceforge.net4Using the Inference Wizard (3)(Example of Cliques in L-S algorithm)Courtesy of Haipeng Guohttp://bndev.sourceforge ...

Subjects

Informations

Published by
Reads 20
Language English
Report a problem
Contents
BNJ 2.03a Beginner Developer Tutorial
Chris H. Meyer (revised by William H. Hsu) Kansas State University KDD Laboratory http://www.kddresearch.org http://bndev.sourceforge.net
n Introduction n Inference Tutorial n Learning Tutorial n Coding the Wizards
http://bndev.sourceforge.net
1
BNJ 2.0a Tools
n Offers many new tools in GUI form n This lecture will focus on the Inference and Learning Wizards n We will also look at components such as evidence, CPT tables, and algorithms behind learning
Contents
http://bndev.sourceforge.net
n Introduction n Inference Tutorial n Learning Tutorial n Coding the Wizards
http://bndev.sourceforge.net
2
Starting the Inference Wizard (1)
n Select Tools à Inference Wizard
http://bndev.sourceforge.net
Starting the Inference Wizard (2) n Load existing network or GUI network
n You may also select to have an evidence file present
http://bndev.sourceforge.net
3
Using the Inference Wizard (1)
n Exact Inference Methods ¨ LS / Junction Tree ¨ Variable Elimination (elimbel) ¨ Loop Cutset Conditioning ¨ Pearl’s Propagation (tree only)
http://bndev.sourceforge.net
Using the Inference Wizard (2)
n L-S Algorithm contains 2 main steps: ¨ Creates a tree of cliques (junction tree) from the Bayesian Network
¨ Computes probability of cliques, then single-node properties are formed based on probability of cliques
http://bndev.sourceforge.net
4
Using the Inference Wizard (3)
(Example of Cliques in L-S algorithm) Courtesy of Haipeng Guo http://bndev.sourceforge.net
Using the Inference Wizard (4)
n Variable Elimination ¨ Uses confactors and the VE algorithm instead of trees n Loop Cutset ¨ Finds minimum cutsets of probability in Bayesian Networks and computes probability from the cutsets
http://bndev.sourceforge.net
5
Using the Inference Wizard (5)
n Pearl’s Propagation ¨ Uses message-passing as data from 1 vertex propagates to all neighbors, then to neighbor’s neighbors, etc … n All algorithms are useful in the correct circumstance, but full explanation is not in the scope of this lecture
http://bndev.sourceforge.net
Using the Inference Wizard (6)
n Approximate Inference ¨ Logic Sampling, Forward Sampling, Likelihood Weighting, Self-Importance Sampling, Adaptive-Importance Sampling, Pearl MCMC Method, Chavez MCMC Method n Again, all algorithms are useful in the correct circumstance n Output your data to chosen file on completion
http://bndev.sourceforge.net
6
Contents
n Introduction n Inference Tutorial n Learning Tutorial n Coding the Wizards
http://bndev.sourceforge.net
Starting the Learning Wizard (1)
n Select Tools à Learning Wizard
http://bndev.sourceforge.net
7
Starting the Learning Wizard (2) n Load Local File Or Database File
http://bndev.sourceforge.net
Using the Learning Wizard (1)
n Select your Learning Algorithm ¨ K2, Genetic Algorithm Wrapper for K2 (GAWK), Genetic Algorithm on Structure, Greedy Structure Learning, Standard Hill-Climbing, Hill-Climbing with adversarial reweighting, Hill-Climbing with Dirichlet prior, Simulated Annealing, Stochastic structural learning
http://bndev.sourceforge.net
8
Using the Learning Wizard (2)
n Depending on user’s desire, different learning algorithms will prove to be more effective
n Output results to file with desired ordering
Contents
http://bndev.sourceforge.net
n Introduction n Inference Tutorial n Learning Tutorial n Coding the Wizards
http://bndev.sourceforge.net
9
Inference Wizard Coding (1)
n GUI ¨ Main GUI window passed as owner of InferenceWizard.java
¨ All Buttons, JButtons, JRadioButtons, etc. are added to InferenceWizard.java’s ActionListener(actionEvent e) method
http://bndev.sourceforge.net
Inference Wizard Coding (2)
n Secondary Windows ¨ Built using BNJFileDialogFactory class
n GUI = First 500 lines of code for InferenceWizard.java
n Main Brain = Last 100 lines of code
http://bndev.sourceforge.net
10
Inference Wizard Coding (3)
n Approximate Inference is slightly more complicated à
http://bnd
Inference Wizard Coding (4)
n JTree , DefaultTreeModel , and InferenceResult are used for inference calculation
n These are found in javax.swing.tree.* and edu.ksu.cis.bnj.bbn.inference.*
http://bndev.sourceforge.net
11